3
$\begingroup$

Is there a methodology to select a reasonable threshold for copy number variation in a CNV (SNP array) TCGA dataset, to define when there is a significative alteration?

Can I download CNV data for normal samples and take the 95th percentile of distribution? Are there better methods?

Update

This is the percentiles plot of the two distribution (Tumor vs Normal) of values, for the same technology (SNP array) and the same genome (hg19).

The tumor distribution has slightly more extreme values, though it is not enough in my opinion. For this reason I think I should not use a percentile score (the 5th and the 95th percentile of the normal samples distribution, for instance) to define the thresholds to call CNV alterations in the tumor samples.

distributions

$\endgroup$
1
$\begingroup$

Your suggested approach of comparing to the baseline distribution on a point-by-point basis isn't bad, although it's going to be susceptible to small false positives from noise. You'd probably want to only use events that span a certain minimum number of consecutive observations.

You might also want to look into circular binary segmentation, as described here: http://www.ncbi.nlm.nih.gov/pubmed/15475419.

$\endgroup$
  • $\begingroup$ I have updated my question, could you please give a look at it and let me know what do you think? $\endgroup$ – gc5 Jun 26 '15 at 15:11
  • $\begingroup$ Hmm, okay, I think I see. No, you don't want to compare the intensity of the tumor at any one position against the 5-95% distribution of the normal across the whole array. You want to compare the intensity of the tumor at one position against the normal at that position. This will normalize for sequence-specific binding differences. $\endgroup$ – Transcriptase Jun 29 '15 at 12:50
1
$\begingroup$

There are many methods for the analysis of CNV. If you are an R user I would recommend you to take a look at the Bioconductor package list, in particular the section for copy number variation. Currently it contains 50 packages!

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.